Blind identification of MIMO discrete‐time systems based on independent component analysis

This paper proposes a blind identification algorithm of a multivariate autoregressive moving average model based on independence of source signals. In such a model, it is difficult to perform blind identification due to nonlinear regression problems. This difficulty is overcome by introducing a feedback structure into the independent component analysis. By using a polynomial matrix fraction, coefficients matrices are exactly identified whereas a conventional method gives mere approximate results. The effectiveness of the proposed method has been confirmed through numerical simulations. © 2007 Wiley Periodicals, Inc. Electron Comm Jpn Pt 2, 90(11): 17–25, 2007; Published online in Wiley InterScience (www.interscience.wiley.com). DOI 10.1002/ecjb.20418